Xiaoyang Zhang1 Mark Friedl2 Science Review Panel Meeting Biosphere 2, Tucson, AZ - January 4-5, 2011 Vegetation Phenology and Vegetation Index Products from Multiple Long Term Satellite Data Records Phenology Science Algorithm Xiaoyang Zhang1 Mark Friedl2 1ERT at NOAA/NESDIS/STAR 2Department of Geography and Environment, Boston University NASA MEASURES #NNX08AT05A

Outline Background Algorithms for developing the MEASURES global land surface vegetation phenology (LSVP) product Preprocessing temporal vegetation index (VI) and determining VI value of background surface Fitting time series vegetative VI trajectory and detecting vegetation phenological metrics Assessing MEASURES global LSVP quality --precision and confidence of phenological product Summary -2- MEASURES VIP ESDRs Science Review Panel

Background

Seasonal Canopy Variation in Deciduous Forests From Baldocchi et al., 2001. -4- MEASURES VIP ESDRs Science Review Panel

Temporal Vegetation Index from Satellite and its Corresponding Phenological Events Greenup Onset Maturity Onset Senescence Onset Dormancy Onset Greenup Phase Senescent Phase Maturity Phase Dormant Phase Maximum VI VI at greenup on set VI increase rate VI decrease rate Growing season length -5- MEASURES VIP ESDRs Science Review Panel

Biophysically Understanding Temporal Trajectory of Satellite Vegetation Index (VI) in Greenup Phase – an example Plant growing (VIv) VIVIb VI=VIv+VIb VI Time Full snow cover (VIs) Partialy snow cover (snow melting) Background (VIb) Plant growing (VIv) VIVIs VIs=0 if without snow cover VI=VIs+VIb VIs=0 if without snow cover VIVIb VI=VIv+VIb VI Time VIa-atmospheric/cloudy contribution +VIa Background -6- MEASURES VIP ESDRs Science Review Panel

Algorithms for Developing the MEASURES Land Surface Vegetation Phenology (LSVP) Product LTDR AVHRR -7- MEASURES VIP ESDRs Science Review Panel

Preprocessing Time Series VI and Determining VI Value of Background Surface -8- MEASURES VIP ESDRs Science Review Panel

LTDR AVHRR QA in NA QA0- good quality QA1-dark vegetation QA2- partially cloudy QA3- cloudy shadow QA4- partially cloudy & cloudy shadow QA5-dark vegetation & partially cloudy QA6- dark vegetation & cloudy shadow QA7-dark vegetation & partially cloudy & cloudy shadow QA8- cloudy QA9-unknown poor quality situation, not processed here. -9- MEASURES VIP ESDRs Science Review Panel

Temporal LTDR AVHRR Data (3-day Composite) for Phenology Detection Note all data except the QA3, QA8, QA9, and QA11 are used for phenology detection because of the limited QA0 data in the time series. Note that QA11 represents snow flag that obtains from snow data. -10- MEASURES VIP ESDRs Science Review Panel

Reconstruction of Time Series Vegetative VI Trajectory (1) Identify background VI value (VIb) Using snow flag and winter land surface temperature (<278K) to determine winter period and background VI (based on data in every two winters) -11- MEASURES VIP ESDRs Science Review Panel

Determination of Background VI Value AVHRR NDVI Land Surface Temperature LST NDVI (x10000) LST Average maxima NDVI during two winters (LST <278K) -12- MEASURES VIP ESDRs Science Review Panel

Ancillary Data for the Determination of Background VI: Snow Cover Jan. 2001 Example: GVIx snow flag in NA (a) NOAA GVIx snow flag (7 day 4 km), global coverage (b) AVHRR daily snow cover inNorthern Hemisphere (1982-2004)(Zhao, H. and Fernandes, R., 2009) The up left pixel is at 90N, 180W. The up right pixel is at 90N, 180E The down left pixel is at 29.5N, 180W. The down right pixel is at 29.5N, 180E. (c) Snow detection from LTDR (U of Arizona) -13- MEASURES VIP ESDRs Science Review Panel

Ancillary Data for the Determination of Background VI: LST LST for the determination of winter period: (1) NCEP North America Regional Reanalysis (NARR) (Mesinger et al., 2006) from 3-hourly LST data at a spatial resolution of 32 km (2) Global 8-day LST averaged from Terra and Aqua MODIS LST from 2003 to 2009 -14- MEASURES VIP ESDRs Science Review Panel

Average Background AVHRR NDVI in NA -15- MEASURES VIP ESDRs Science Review Panel

Reconstruction of Time Series Vegetative VI Trajectory (2) Remove atmospheric/cloud contamination (VIa) Removal of large spurious values Large gap determination and filling using ecosystem rules: Forests: Span between two cycle peaks NOT less than 4 months Shrubland/grassland: Span between two cycle peaks NOT less than 2 months Cropland: Span between two peaks NOT less than 1.5 months Removal of irregular local small values Time series smoothing using Savitzky_Golay filter -16- MEASURES VIP ESDRs Science Review Panel

Removal of Pseudo Plant Cycles NOT two peaks Ecosystem rules: Forests: Span between peaks >4 month Shrubland/grassland: Span between peaks >2 months Cropland: Span between peaks > 1.5 months -17- MEASURES VIP ESDRs Science Review Panel

Fitting Time Series Vegetative VI Trajectory and Identifying Vegetation Phenological Metrics -18- MEASURES VIP ESDRs Science Review Panel

Determination of NDVI Increase and Decrease Sections and Growing Cycles: Fourier Algorithm Fourier algorithm was tried to determine the cycles of vegetation growth. While it works well in deciduous forests, it fails in shrublands with one large cycle and one small cycle. -19- MEASURES VIP ESDRs Science Review Panel

Determination of NDVI Increase and Decrease Trajectories and Vegetation Growing Cycles: Local Moving Slope Algorithm Moving slope is very flexible and is capable of identifying growing cycles from complex temporal trajectories -20- MEASURES VIP ESDRs Science Review Panel

Fitting Temporal Vegetative Trajectory Using Pieceswise Logistic Model Varying a and b Greenup phase t is time in days VI(t) is the VI value at time t VIv is the VI value in vegetation growth a and b are vegetation growth parameters c+VIb is the maximum VI value VIb is the background VI value Varying c Greenup phase Zhang et al., 2003 -21- MEASURES VIP ESDRs Science Review Panel

Types of Temporal Trajectories in Vegetation Index Across Globe -22- MEASURES VIP ESDRs Science Review Panel

Identifying Phenological Transition Dates The rate of curvature change -23- MEASURES VIP ESDRs Science Review Panel

Temporal Trajectory of AVHRR NDVI and Field Observations at Harvard Forest. -24- MEASURES VIP ESDRs Science Review Panel

Greenup Onset from AVHRR in 1996 b Recorded in the first cycle Recorded in the second cycle The first greenup onset for the year (combined from the first and the second cycles) c

Dormancy Onset from AVHRR in 1996 b a Recorded in first cycle Recorded in second cycle The first greenup onset for the year (combined from the first and the second cycles) c

Assessment of MEASURES Global LSVP Quality --Precision and Confidence of Phenological Product (for both AVHRR and MODIS/VIIRS) -27- MEASURES VIP ESDRs Science Review Panel

Assessment of Fitted Temporal Vegetative VI Trajectory Quality of curve fitting using Willmott’s index of agreement (Willmott, 1982 ): where n is number of observations with good (or better) quality during vegetation growing season P(i) are the fitted values O(i) are the observations with good quality Ō is the mean observed value. This index of agreement provides a measure of relative error in model estimates -28- MEASURES VIP ESDRs Science Review Panel

Assessment of Good Quality Observations in the Time Series Proportion of good quality observations during a growing season: Pqa0=Nqa0/T T—is the total number of 3-day VI during a growing season Nqa0—is the number of good quality observation within a growing season. If there is one good value within a moving window of 3 3-day VI, it is counts as one good observation. This is due to the fact that the error in phenology detection is lowest at temporal resolution of 6 or 8 days (Zhang et al., 2009). Zhang et al., 2009 -29- MEASURES VIP ESDRs Science Review Panel

Determination of Quality in Phenology Detection by Combining the Willmott’s Index of Agreement and the Proportion of Good Quality Data in Satellite Observations Willmott’s index of agreement 1996 Proportion of good quality observations 1996 Quality assurance 1996 QA=0 (processed, good quality) if Pqa0 >0.6 and d>0.8 QA=1 (processed, other quality) if Pqa0 >0.3 and d>0.8 QA=2 (not processed, clod) if Pqa0 <0.3 QA=4 (not processed, other ) if growing season amplitude in VI<0. 08 in forests and <0.02 in other ecosystems

Conclusions Reconstruction of vegetative seasonal trajectory (including the identification of background VI value, the removal of atmospheric/cloud contamination, and the determination of growing cycles) is the most important part in phenology detection. Pieceswise logistical model is robust in fitting various complex trajectories of vegetation growth. It’s curvature change rate is effective for the identification of phenological transition dates and other metrics. The quality of phenological detection could be assessed using the combination of good quality satellite observations during a vegetation growing season and the quality of curve fitting (quantified using Willmott’s index of agreement). The developed algorithm is tested using AVHRR data which are of relatively poor quality. The algorithm would produce better phenology detections from MODIS/VIIRS data which are high quality. -31- MEASURES VIP ESDRs Science Review Panel

THANKS Next : ESDR Specification

ESDR specification

Land Surface Vegetation Phenology Science Data Sets (SDS) DataField Name Format ------------------------------------------------------------------------------------------ DataField_1 Onset_Greenness_Increase UINT16 DataField_2 Onset_Greenness_Maximum UINT16 DataField_3 Onset_Greenness_Decrease UINT16 DataField_4 Onset_Greenness_Minimum UINT16 DataField_5 VI_Onset_Greenness_Minimum UINT16 DataField_6 VI_Onset_Greenness_Maximum UINT16 DataField_7 VI_Area UINT16 DataField_8 Growing_Season_Length UINT16 DataField_9 Rate_Greenness_Increase UINT16 DataField_10 Rate_Greenness_Decrease UINT16 DataField_11 Dynamics_QC UINT16 -34- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification Product name convention LSP12C2.A1982001_1982365.001.2010266161623.hdf LSP12C2.A: product identification 1982001_1982365: time period of vegetation phenology detection 001: the data product version 2010266161623: time of product processing .hdf: the output file is in HDF --------------- -35- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) Grid Structure: LSP_Grid_LSP Dimensions: Dimension Dimension Name Value Dimension_1 Ydim:MGLSP_Grid_LSP Data Rows Dimension_2 Xdim:MGLSP_Grid_LSP Data Columns Dimension_3 Num_QC_Words:MGLSP_Grid_LSP Num_QC_Words Dimension_4 Num_Modes:MGLSP_Grid_LSP Num_Modes global attributes: :WestBoundingCoordinate = -180.0 ; :EastBoundingCoordinate = 180.0 ; :NorthBoundingCoordinate = 90.0 ; :SouthBoundingCoordinate = -90.0 ; :PixelSize = 0.05deg ; -36- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) Format for SDS of Onset Greenness Increase, Onset Greenness Maximum, Onset Greenness Decrease, and Onset Greenness Minimum DataField Name Data Type Dimensions DataField_1 Onset_Greenness_Increase UINT16 Dimension_1 Dimension_2 Dimension_3 Description: Onset of greenness Increase --- Days starting from January 1, 1980 Note: Word 1 is first Mode of the year, Word 2 is second mode of the year (other possible modes are not reported) Data conversions: DOY=file data - (given year-1979)*366 DataField_1 HDF Attributes: long_name STRING 1 PGE "Onset_Greenness_Increase" units STRING 1 PGE "day" valid_range uint16 2 PGE 1, 32766 _FillValue uint16 1 PGE 32767 scale_factor float64 1 PGE 1 add_offset float64 1 PGE 0 -37- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) Format for SDS of VI Onset Greenness Increase and VI Onset Greenness Maximum DataField_5 VI_Onset_Greenness_Increase UINT16 Dimension_1 Dimension_2 Dimension_3 Description: VI value at onset of greenness increase during a growth cycle Note: Word 1 is first Mode of year, Word 2 is second mode of year (other possible modes are not reported DataField_5 HDF Attributes: long_name STRING 1 PGE "VI_Onset_Greenness_Increase" units STRING 1 PGE "VI value" valid_range uint16 2 PGE 0, 10000 _FillValue uint16 1 PGE 32767 scale_factor float64 1 PGE 0.0001 add_offset float64 1 PGE 0 -38- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) DataField_8 Growing_Season_Length UINT16 Dimension_1 Dimension_2 Dimension_3 Description: Number of days in a growing cycle Note: Word 1 is first Mode of year, Word 2 is second mode of year (other possible modes are not reported DataField_7 HDF Attributes: long_name STRING 1 PGE "Growing_Season_Length" units STRING 1 PGE "day" valid_range uint16 2 PGE 0, 32766 _FillValue uint16 1 PGE 32767 scale_factor float64 1 PGE 1 add_offset float64 1 PGE 0 -39- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) Format for SDS of greenness increase rate and greenness decrease rate DataField_9 Rate_Greenness_Increase UINT16 Dimension_1 Dimension_2 Dimension_3 Description: Average rate of VI increase during a greenup phase Note: Word 1 is first Mode of year, Word 2 is second mode of year (other possible modes are not reported DataField_9 HDF Attributes: long_name STRING 1 PGE "Rate_Greenness_Increase" units STRING 1 PGE "VI/day" valid_range uint16 2 PGE 0, 32766 _FillValue uint16 1 PGE 32767 scale_factor float64 1 PGE 0.0001 add_offset float64 1 PGE 0 -40- MEASURES VIP ESDRs Science Review Panel

Land Surface Vegetation Phenology Product Specification (continue) DataField_11 Dynamics_QC UINT16 Dimension_1 Dimension_2 Description: Quality flags for vegetation phenology DataField_11 HDF Attributes: Note: First Word: the first two bits are Mandatory QA 0=processed, good qual 1=processed, other qual 2=not processed, cloud 3=not processed, other the next two bits are TBD the 5-8 bits are Land Water mask (as passed down from NBARS) 0 = Shallow ocean 1 = Land (Nothing else but land) 2 = Ocean coastlines and lake shorelines 3 = Shallow inland water 4 = Ephemeral water 5 = Deep inland water 6 = Moderate or continental ocean 7 = Deep ocean the 9-16 bits are Phenology__Assessment 0-100 = Assessment values 255= FillValue -41- MEASURES VIP ESDRs Science Review Panel

Backup slides

AVHRR NDVI Quality

3-day VI Time Series from Daily LTDR AVHRR Data for the Development of MEASURES LSVP Products The priority in the generation of 3-day time series is: QA0-good quality QA1-dark vegetation QA2-partially cloudy  QA4-partially cloudy & cloudy shadow  QA5-dark vegetation & partially cloudy QA6-dark vegetation & cloudy shadow  QA7-dark vegetation & partially cloudy & cloudy shadow QA3-cloudy shadow QA8- cloudy  QA9-unknown poor quality situation, not processed In the composite, the highest VI is selected if the QA is the same for a 3-day period

Maturity Onset from AVHRR in 1996 b Recorded in the first cycle Recorded in the second cycle The first greenup onset for the year (combined from the first and the second cycles) c

Senescent Onset from AVHRR in 1996 b Recorded in the first cycle Recorded in the second cycle The first greenup onset for the year (combined from the first and the second cycles) c